Mutually Unbiased Bases Are Complex Projective 2-Designs
Total Page:16
File Type:pdf, Size:1020Kb
Mutually Unbiased Bases are Complex Projective 2-Designs Andreas Klappenecker Martin R¨otteler Department of Computer Science NEC Labs America, Inc. Texas A&M University 4 Independence Way College Station, TX, 77843–3112, USA Princeton, NJ 08540 U.S.A. Email: [email protected] Email: [email protected] Abstract— Mutually unbiased bases (MUBs) are a primitive that we want to determine the density matrix ρ of an ensemble used in quantum information processing to capture the principle of quantum systems using as few non-degenerate observables of complementarity. While constructions of maximal sets of d+1 as possible. We assume that it is possible to make a com- such bases are known for system of prime power dimension d, itis plete measurement of each observable O = x b b , unknown whether this bound can be achieved for any non-prime b∈B b | ih | power dimension. In this paper we demonstrate that maximal sets meaning that the statistics tr(ρ b b )= b ρ Pb is known for | ih | h | | i of MUBs come with a rich combinatorial structure by showing each eigenvalue xb in the spectral decomposition. Ivanovi´c that they actually are the same objects as the complex projective showed in [12] that complete measurements of at least d +1 2-designs with angle set {0, 1/d}. We also give a new and observables are needed to reconstruct the density matrix. He simple proof that symmetric informationally complete POVMs are complex projective 2-designs with angle set {1/(d+1)}. also showed that this lower bound is attained when d+1 non- degenerate pairwise complementary observables are used. I. INTRODUCTION A simple example is provided by the Pauli spin matrices σx, Two quantum mechanical observables are called comple- σy, σz. A complete measurement of these three observables mentary if and only if precise knowledge of one of them allows to reconstruct a 2 2 density matrix, a fact appar- implies that all possible outcomes are equally probable when ently known to Schwinger× [18]. Nowadays, we know how to measuring the other, see for example [19, p. 561]. The do this state tomography process—at least in principle—in principle of complementarity was introduced by Bohr [6] in dimensions d = 3, 4, and 5. It is an open problem whether 1928, and it had a profound impact on the further development it is possible to perform this kind of state tomography in of quantum mechanics. A recent application is the quantum dimension 6, because the construction of a set of 7 mutually key exchange protocol by Bennett and Brassard [3] that unbiased bases in dimension d =6 is elusive. exploits complementarity to secure the key exchange against eavesdropping. II. MUTUALLY UNBIASED BASES We mention a simple mathematical consequence of this d complementarity principle, which motivates some key notion. Definition 1: Two orthonormal bases B and C of C are ′ called mutually unbiased iff b c 2 =1/d holds for all b B Suppose that O and O are two hermitian d d matrices |h | i| ∈ representing a pair of complementary observables.× We assume and c C. arXiv:quant-ph/0502031v2 11 Feb 2005 ∈ that the eigenvalues of both matrices are multiplicity free. The goal is to construct d + 1 mutually unbiased bases It follows that the observables O and O′ respectively have (MUBs) in any dimension d 2. There are several con- ≥ orthonormal eigenbases B and B′ with basis vectors uniquely structions known to obtain MUBs. At least for prime power determined up to a scalar factor. dimension the problem is completely solved. This follows The complementarity of O and O′ implies that if a quantum from Constructions I-III below. However, in any dimension system is prepared in an eigenstate b′ of the observable O′, other than a prime power it is unknown if a maximal set and O is subsequently measured, then the probability to find of d + 1 MUBs can be found. The best known result is the system after the measurement in the state b B is given Construction IV below which only works in dimensions d by b b′ 2 =1/d. Recall that two orthonormal∈ bases B and which are squares and never gives a maximal set of MUBs. ′ |h |Ci|d B of are said to be mutually unbiased precisely when Construction I (Wootters and Fields [24]) Let q be an odd ′ 2 ′ ′ b b = 1/d holds for all b B and b B . Thus prime power. Define the|h | eigenbasesi| of non-degenerate complementary∈ ∈ observables 2 are mutually unbiased. Conversely, we can associate to a −1/2 tr(ax +bx) Cq va,b = q (ωp )x∈Fq , pair of mutually unbiased bases a pair of non-degenerate | i ∈ complementary observables. with ωp = exp(2πi/p). Then the standard basis together with There is a fundamental property of mutually unbiased bases the bases Ba = va,b b Fq , a Fq, form a set of q +1 that is invaluable in quantum information processing. Suppose mutually unbiased{| basesi | of∈ Cq}. ∈ Construction II (Galois Rings [13]) Let GR(4,n) be a finite Then we have that: Galois ring with Teichm¨uller set n. Define • M(pr)= pr +1 for p prime, r N, T ∈ 2πi • M(n) n +1 for all n N, v =2−n/2 exp tr(a +2b)x . ≤ ∈ N a,b 4 • M(mn) min M(m),M(n) for all m,n . | i x∈Tn ≥ { } ∈ • M(d2) N(d), where N(d) is the number of mutually ≥ Then the standard basis together with the bases Ma = orthogonal Latin squares of size d d. n va,b b n , a n, form a set of 2 + 1 mutually × n An open problem is to show that lim inf n→∞ M(n)= . unbiased{| i | bases∈ T of} C2∈. T ∞ Construction III (Bandyopadhyay et al. [1]) Suppose there III. WELCH’S LOWER BOUNDS exist subsets 1,..., m of a unitary error basis such that Suppose that X is a finite nonempty set of vectors of unit C C B Cd i = d, i j = 1d for i = j, and the elements of i norm in the complex vector space . The vectors in X satisfy pairwise|C | commute.C ∩C Let{ M} be a matrix6 which diagonalizes C. the inequalities i Ci Then M1,...,Mm are MUBs. 1 1 x y 2k , (1) Construction IV (Wocjan and Beth [23]) Suppose there are X 2 |h | i| ≥ d+k−1 | | x,yX∈X k w mutually orthogonal Latin squares [4], each of size d d over the symbol set S = 1,...,d . Then w + 2 MUBs× for all integers k 0. Welch derived these bounds in [22] ≥ in dimension d2 can be constructed{ } as follows. With each to obtain a lower bound on the maximal cross-correlation of Latin square L (and additionally the square (1, 2,...,n)t spreading sequences of synchronous code-division multiple- (1,..., 1)) we can associate vectors of length d over the⊗ access systems. Blichfeld [5] and Sidelnikov [20] derived alphabet 1,...,d2 : for each symbol α S define a vector similar bounds for real vectors of unit norm. {d } ∈ sL,α C as follows: start with the empty list sL,α = . A set X attaining the Welch bound (1) for k = 1 is Then∈ traverse the elements of L column-wise starting at the∅ called a WBE-sequence set, a notion popularized by Massey upper left corner. Whenever α occurs in position (i, j) in L, and Mittelholzer [15] and others. Using equation (1), it is then append the number i + jd to the list sL,α. The other straightforward to check that the union of d + 1 mutually ingredient to construct these MUBs is an arbitrary complex unbiased bases of Cd form a WBE-sequence set. These Hadamard matrix H = (hi,j ) of size d d. For each Latin extremal sets of mutually unbiased bases are even better, since square L and each α, j 1,...,d define× a normalized they also attain the Welch bound for k = 2. In fact, we √ ∈d { } show that a sequence set attains the Welch bounds (1) for all vector vL,α,j := 1/ d i=1 esL,α[i]hi,j , where ei are the | i 2 elementary basis vectorsP in Cd . Then the bases given by k t if and only if it is a t-design in the complex projective space≤ CP d−1. BL := vL,α,j : α, j = 1,...,d together with the identity {| i } Let us introduce some notation. Let Sd−1 denote the sphere matrix 1d2 form a set of w +2 MUBs. of unit vectors in the complex vector space Cd. We say that Example 1: In dimension d = 3 Construction I yields the two vectors u and v of Sd−1 are equivalent, in signs u bases v, if and only if u = eiθv for some θ R. It is easy to≡ −1/2 2 2 ∈ 3 (1, 1, 1), (1,ω3,ω3), (1,ω3,ω3) , see that is an equivalence relation. We denote the quotient { } ≡ − − − −1/2 2 2 manifold Sd 1/ by CSd 1. Notice that the manifold CSd 1 3 (1,ω3,ω3), (1,ω3, 1), (1, 1,ω3) , { } is isomorphic to≡ the complex projective space CP d−1, but we 3−1/2 (1,ω2,ω2), (1,ω , 1), (1, 1,ω ) , { 3 3 3 3 } prefer the former notation because normalizing vectors to unit which together with the standard basis 13 form a maximal length is common practice in quantum computing.